Finite Element Modeling and Analysis Dos and Donts

Hormoz Zareh, Associate professor Mechanical Engineering Dept. Portland State University SAE Meeting, October 24, 2003

What is Finite Element Analysis or Modeling?

Analysis: A numerical method for approximation of a
complex differential equation problem (discrete or continuous) into a finite number of solvable algebraic equations.

Modeling: Subdividing a physical structure into a number

of small (finite) entities (elements). This is followed by application of prescribed boundary conditions (loads and restraints).

What is F.E. Analysis all about?

Fundamental design question:

What type of analysis is needed, and how will the

outcome be related to the design performance requirements?

Requirements to perform FEA/FEM

Grasp of the problem area
Knowledge of the field (Structural, thermal, etc.) Ability to solve a simplified version via analytical methods

Fundamental background in FE modeling and Analysis

Take formal training courses, not just vendors how to class. Think about all the loading conditions that could affect the design, including thermal, mechanical, motion, etc. Simplify CAD models to the minimum required detail. Start with a simple model, and add complexity later. Double check the input data (material and physical properties). Examine the model with a range of boundary conditions instead of just one. Indicates design sensitivities! Discuss the model and results with your colleagues.

More about Fundamental requirements

Remember that FE modeling is more than preparing a mesh
and preprocessing.

Skill in modeling is based on the visualization of physical

behavior and relating it to the specific element behavior.

Acquire broad knowledge base. Know and remember the assumptions and underlying
restrictions of the analysis tools.

Always start simple and be prepared to revise your model.

Evaluate the results with a healthy dose of skepticism.

Structure behavior vs. Element behavior

The single most important underlying question in FE
modeling and analysis:

Which element should I use for a given problem?

The answer is not simple.

Correct element identification, mesh density, and application

of boundary conditions are all part of the following principle: Thorough understanding of how the original design is likely to behave under the actual loads/boundary conditions, and how selected elements are able to simulate that behavior

Some general Modeling guidelines:

Include as much of the structure as possible in the model.
Do not omit part of the structure on the assumption that it
does not influence the rest of the system by being lightly stressed!

Use finer mesh (than the one used to obtain

displacements) to obtain stresses.

Use symmetry when appropriate, avoid when unsure about

the outcome.

General Modeling guidelines:

(Continued)

Keep the design detail at a level which can be easily

managed for future analysis and verification.

Generate the beam models with curves and edges and not
the actual cross sections.

Eliminate minor geometric details such as fillets, chamfers,

tiny holes, etc. They complicate the FE mesh while not influencing the analysis outcome.

General Modeling guidelines:

(Continued)

If some details are necessary, include them such that

they can be easily identified and suppressed prior to FE modeling and analysis.

Take advantage of partitioning to divide complex shapes

into a series of simpler geometries.

General Modeling guidelines:

(Continued)

If the problem involves nonlinearity or anisotropy, analyze

a linear isotropic version of the problem first.

When dynamic effects are present, perform a static

analysis first, using loads that approximate the dynamic effect.

Do not use a symmetric half of the model for modal

analysis (Natural frequencies and modes).

Sources of Error and Concept of Error estimation

One must distinguish between inherent errors in the FE
process and plain mistakes.

The mistakes include such things as entering the wrong data

(such as material or properties, thickness of a member, forgetting the support or loads, etc.)

Modeling errors include using the wrong element

formulation, modeling the incorrect geometry, using a poor mesh distribution, and numerical error (due to limitation of computing accuracy).

Sources of Error and Concept of Error estimation

(Continued)

Another source of modeling error may arise when a load is

distributed over a small region.

The solution at the immediate vicinity of the load requires

considerable mesh refinement. However, if the desired area is a reasonably far away from the load location, coarse mesh could be sufficient.

Rule of thumb based on St. Venants principle:

The error is generally small when the distance (at which measurement is taken) is at least five times the width of the original distributed load.

251-elem. mesh, Max. stress = 3540 954-elem. mesh, Max. stress = 4370

No sign of convergence and one should not be expected!

Element-by-element stress field

This type of stress distribution is discontinuous between

elements, while stresses computed from nodal averaging are continuous.

One method of judging an approximate error field is to

observe the difference between element-by-element and nodal averaging quantities.

Typically, the amount of discontinuity between elements is

regarded as a measure of error.

For example: Consider the stress contour indicating stress variation within a single element as follows:
Contour
1 2 3 4

Stress [Psi]
2511 2970 3380 3820

The average stress within the element is 3170 with an estimated highest error of almost 21% !

largest error = (3170 2511) / 3170 = 0.21 or 21%

Mesh refinement
This is based on the assumption that discontinuities in
element-by-element stress field are indicative of error. However, physical discontinuities may exist which produce realistic, and often desirable, stress discontinuities.

Mesh should not be refined to the extent that the size of the
smallest element is considerably smaller than the largest one.

Finally, adaptive meshing may give the analyst a false sense

of security that by achieving convergence, a correct solution has been obtained.

Rule of thumb: Keep the ratio to below 100:1 (linear

dimensions) or 1000:1 (area dimensions).

Element Selection
Forms the foundation of FE modeling and analysis. Example: 4-node QUADS are preferred to 3-node triangular
elements.

Linear formulations with successive refinements yield good

results.

Avoid higher order elements unless necessary, or in critical

situations (medical device, aerospace).

Elements
Meshing examples:

Example of poor mesh transition (Too abrupt from coarse to fine)

Example of better mesh transition (smoother transition from coarse to fine)

Elements (continued)
More examples

Examples of poorly shaped 2-D elements

Elements (continued)
Linear (4-node) Tetrahedron Quadratic (10-node)

Fine for thermal analysis, Bad for structural analysis

Very good choice for thermal as well as structural analysis

Physical and Material properties

Make sure proper material and physical properties are

used in the analysis.

A common mistake is to inadvertently use the softwares

defaults for the analysis.

Some default properties used in certain FEA program:

Thickness of Shell elements = 0.038 in. (1 mm) Material property = Generic Isotropic Steel

Boundary Conditions
A critical aspect of the modeling is the proper
transformation of the actual boundary conditions to the loads/restraints set of the FE software. For example, Dynamic (Modal) analysis is very sensitive to the boundary conditions.

A model must be fully restrained for proper analysis.

Otherwise, RIGID BODY modes will cause solution failure (in Static analysis), or unexpected vibration modes (in Dynamic analysis).

Boundary Conditions (continued)

For models which are not fully restrained: Static analysis: completely restrain one node. Pick a node
which is not close to the area of deflection/stress which is being investigated.

Dynamic analysis: set the first natural frequency to a small

number greater than zero, yet small enough to not interfere with the actual frequency.

Thermal analysis: restraints are applied as temperature.

Condition of symmetry is imposed as insulation!

Solution strategy

Decide on a solution strategy which best meets the design

requirements. Static analysis vs. Dynamic analysis Modal analysis vs. Shock (impact) loading Structural vs. Thermal (or combined) analysis

First few modes represent displacements quite well, but

impact loading cannot be modeled correctly with Modal analysis solution!

Post processing
View the results with critical eyes.
Dont be impressed by color contours, and what might
seem to be a convincing output.

Common sense and good engineering judgment are far

more important than the computer output.

Model with full detail

Model with details suppressed

Elements = 2 565 Elements = 66 727 No. of equations = 311 421 No. of equations = 14 889

Basics of CMP Machine

Source: http://lma.berkeley.edu/precision/

CMP Model

wafer Retaining Ring

Pad

Sub-layer

Retaining Ring

Gap elements

58 0.00 -2.50 -5.00

60

62

64

66

68

70

72

74

76

Contact pressure (Mpa)

-7.50 -10.00 -12.50 -15.00 -17.50 -20.00 Distance from center of w afer (m m ) 7 PSI 8 psi 3 PSI 5 PSI 9 PSI 10 PSI

Experimental Results

FEA Results with friction, without ret. ring

FEA Results with Retaining Ring Ring Pressure 8 PSI 6.95 7.411 4.612 40.273

Average Contact Pressure Maximum Contact Pressure Minimum Contact Pressure Pressure Non-uniformity

6.837 11.225 3.814 108.395

6.792 12.347 4.96 108.760

Final Thought

Finite Element Analysis makes a good engineer great,

and a bad engineer dangerous !

Robert D. Cook, Professor of Mechanical Engineering University of Wisconsin, Madison